Water samples (400 ml) were taken from 15 different sampling points (Supplementary Figures 1, 2). Sampling was carried out as close as possible to the central part of the river in order to avoid collecting material from the banks. All samples were obtained on the same day. The samples were collected in sterile plastic bottles, transported to the laboratory at 4°C and subsequently stored at −80°C. Finally, animal and Human fecal samples were also used.

First, the samples were filtered through a paper filter to eliminate soil and/or plant contaminants, and subsequently, the bacteria were collected on a 0.22 μm membrane. The bacteria were then resuspended by stirring the filters in 3 ml of sterile H₂O for 1 h, and a subsequent centrifugation at 12,000 rpm for 30 min collected a bacterial pellet. Bacterial DNA was extracted using the NucleoSpin Tissue kit (Qiagen) following the supplier's directions. The DNA was initially resuspended in a volume of 40 μl of sterile H₂O and subsequently diluted to obtain a final concentration of 5 ng/μl. To create a reference of the microbial communities of fecal samples, samples of human, dog, cat, pig, and bovine feces were analyzed. In this case, the bacterial DNA was obtained using the NucleoSpin Soil kit (Macherey-Nagel) starting from 25 mg of material.

A portion of the 16S rRNA was amplified using 5 μl of the extracted DNA in a final reaction volume of 25 μl using Platinum Taq DNA Polymerase High Fidelity (Thermofisher) in accordance with the manufacturer's instructions. The amplifications were performed for 26 cycles of the following amplification profile: 95°C-60 s (denaturation), 55°C-60 s (annealing) and 72°C-60 s (extension). The following primers were used: Pro341F: 5′-CCTACGGGNBGCASCAG-3′ and Pro805R: Rev 5′-GACTACNVGGGTATCTAATCC-3′ (Takahashi et al., 2014). These primers produce an amplicon of ~445 bp (depending on the amplified bacterial species). The overhang adapter (forward overhang: 5′ TCGTCGGCAGCGTCAGATGTGTATAAGAGACAG-locus specific sequence and reverse overhang: 5′ GTCTCGTGGGCTCGGAGATGTGTATAAGAGACAG-locus specific sequence; from BMR Genomics srl-Italy) was added and followed by library construction. The libraries were purified with Beads Amplure XP 0.8X, amplified with Indexes Nextera XT Illumina, and then normalized, mixed, and loaded on Miseq with 2 × 300 bp (paired-end).

The raw sequences were verified and filtered by quality, trimmed by the primers, and fused with Qiime2 v. 8 software (Bolyen et al., 2019). DADA2 (Qiime2) software isolated the OTUs (Operational Taxonomic Unit), whose sequences were compared against Greengenes v. 13-8 database to obtain the taxonomic assignment. A similarity analysis between the species detected in the feces and those ones in the water was performed using the BLAST program, available at NCBI (Altschul et al., 1990).

The interpretation of the results regarding the percentage of the different bacterial species was carried out using Excel 365. PCA (Principal Components Analysis) analysis was performed using StataMp v. 17 software. The statistical evaluation of the different Shannon–Weaver index obtained (SWi) was calculated with the two sample Mann–Whitney test.

The complexity of the bacterial populations was evaluated using the Shannon–Weaver index (SWi; Figure 1A). A higher value was identified at sampling D (8.06), while a lower value was identified at sampling R (4.37). These values are consistent with what has been observed in other watercourses (Cruz et al., 2021; Tee et al., 2021). The trend of the SWi at the different sampling points indicates that the maximum values were near the source of the river, while the minimum values were near the end of the path. The greatest complexity change occurred between points H and I. This fact could be explained by considering that between those two sampling points, the Lambro river passes from a sparsely inhabited area to a highly inhabited area (municipality of Milan) and then crosses an area densely characterized by farms. The median of the SWi values obtained from the A–H samplings, 7.48, was significantly higher (p < 0.001) than the median of the values obtained at the I–R sampling points, which was 4.94 (Figure 1B). A great variety in the bacterial populations of a watercourse is an indication of good health and represents also a main factor that allows the ecosystem to react with adverse phenomena (Loreau et al., 2001; Cardinale, 2011; Zinger et al., 2012). What has been observed supports the hypothesis that the quality of the water of Lambro worsens significantly when it crosses areas with high anthropic activities.

The deterioration of the water quality was also confirmed by observing the number of OTUs identified (Figure 1A). The number of OTUs, standardized on 15,000 sequences, varied from 126 (Q and R sample points) to 607 (G sample point). The average number of OTUs in the first seven withdrawals (from A to G) was approximately double of the average of the last eight withdrawals (from I to R) at 423 compared to 220, respectively.

Considering all 15 sampling points, bacteria belonging to 37 different phyla were identified, even if only 16 of them were present at a percentage >0.5% in at least one sample (Figure 2A). The most represented phyla were Proteobacteria, Bacteroidetes, Firmicutes, and Actinobacteria. They all represent from a minimum of 79.05% (point C) to a maximum of 99.99% (point R) of the entire bacterial community; furthermore, after sampling point D, their percentage was always higher than 90% (Figure 2B). Proteobacteria represented the majority phylum at all sampling points, and their percentage never dropped below 50%. The bacteria belonging to the Bacteroidetes, in contrast, showed an almost constant trend upward to sampling point L and then an increase to the last sampling point. Sampling point L marks the entrance of the Lambro in the region largely characterized by the presence of farms. The Firmicutes exhibited an opposite trend, in fact their presence was appreciable up to point I and then disappeared in the subsequent sampling points, while the Actinobacteria showed a constant presence in all sampling points, except points I and L, where they reached higher percentage. Finally, regarding the other phyla observed (always among those present in at least 0.5% of at least one sampling point), the data are shown in Supplementary Figure 3.

The analysis of the bacterial species identified in the 15 samples revealed that 87 of them showed a presence higher than 1% in at least one single sample (the list of these species is reported in Supplementary Figure 4). Considering the total percentage of these species at sampling points, it is possible to note an increasing trend from the source (60%) to the end of the path (92%; Supplementary Figure 5). An opposite trend could be observed by considering the number of species that possessed a presence higher than 1% at the individual sampling points, which included 15/20 near the source to 8/10 near the last sampling points. In both cases, the variations were not progressive but occurred abruptly between the withdrawal points H and I. The observed result confirms what has already been seen observed previously regarding the sudden change in the quality of the Lambro when crossing it crosses the most populated area (municipality of Milan).

Among all the species observed, only one, Acinetobacter johnsonii, was present in quantities higher than 1% in all 15 samples. This bacterium made up 4.5% of the bacterial community in sample A and 60.7% in sample R; however, in this case, the growth was not gradual and constant but there was a sudden growth between withdrawal H (7.7%) and withdrawal I (54.8). The same trend, albeit with lower percentages, could be observed for an unidentified bacterium belonging to the genus Acinetobacter. Finally, in addition to the two bacteria previously mentioned, only four other bacteria had a presence higher than 10% in at least one of the sampling points: a bacterium of the genus Enhydrobacter (32.2% in sample B), a bacterium of the genus Delftia (24.1% in sample A), a bacterium belonging to the genus Flavobacterium (20.9% in sample R), and, finally, a bacterium of the genus Arcobacter (12.9% in sample H). Acinetobacter johnsonii is a very common bacterium in the environment, so it is not surprising that it was the most common microorganism found (Guardabassi et al., 1999). Moreover, it rarely poses a health hazard. Bacteria of the genus Deftia, and in particular Deftia acidovorans, prefers an environment with low salt concentrations that are typical of high-altitude springs and known for their ability to aggregate gold and produce small nuggets (Johnston et al., 2013).

A further interesting fact emerged from the principal coordinates analysis (PCA; Figure 3). The results clearly revealed two clusters of withdrawal points. The first group was associated with the samples taken before the entrance of the Lambro into highly populated areas (red circle in Figure 3), while the second group included sample points belonging to the region with the highest agricultural density (green area in Figure 3). This result is very interesting as there is a close relationship between the geographic position of the sampling point and the composition of the microbial community. Observing the phylums identified in these two groups and the variations (Supplementary Figure 6) it can be observed that only three phylum undergo a percentage increase (Proteobacteria, Bacteroidetes and Actinobacteria), while all the others undergo a strong decrease. This observation is consistent with what was previously observed regarding the decrease in bacterial variability after sampling point H.

To obtain information about the Bacteroidetes present in feces, samples belonging to five different species were analyzed: human, cattle, pig, dog, and cat. The phyla identified in the five different types of feces are shown in Supplementary Figure 7A. In the samples, most of the microbial components were made up of Firmicutes and Bacterioidetes, which together never comprised less than 90% of the total microbial flora. With regard to the feces samples, the number of OTUs (standardized to 15,000 sequences) and the Shannon–Wiener index are indicated in Supplementary Figure 7B. The composition observed was consistent with what was previously published (Bermingham et al., 2018; Mehta et al., 2018; Lourenco et al., 2020; Pilla and Suchodolski, 2020; Wylensek et al., 2020). By analyzing the species of Bacteroidetes present at a percentage higher than 2.5% in the stool samples, it was possible to confirm the great specificity of this phylum with the origin of the feces (Table 1). Of the 38 species considered, only five were present in the feces of at least two species, and four involved man, dog, and cat. In fact, the species present in pig and bovine feces were strictly specie-specific. The analysis of the similarities between these specie of Bacteroidetes and those identified in the water withdrawals indicated that there were only six similarities higher than 99% (Table 2). The results show that a porcine-specific Bacteroidetes (SS4, 3.8%) was identified at a percentage higher than 0.1% at sampling point C. Furthermore, at the same sampling point, a specie was found that had a high similarity with another specific specie of the pig (SS1, 8.4%). Finally, it was noted that at the sampling points M, N and Q, a bacterial specie very similar to Bacteroidetes specific to man (HS3, 6.2) and cat (FC10, 3.6) was identified. Sampling point C corresponds to a municipality where it is possible to hypothesize the presence of family-type pig farms.